Indoor scene generation method, device and computer equipment based on scene boundary

Abstract

The present application relates to a method, device and computer equipment for generating indoor scenes based on scene boundaries. The method comprises: obtaining the scene boundary and the current indoor objects of the current indoor scene, predicting the newly added objects in the current indoor scene based on the scene boundary and the current indoor objects, and obtaining the currently newly added objects; determining the currently associated objects associated with the currently newly added objects in the current indoor objects based on the object layout features corresponding to the current indoor objects, and obtaining the local scene boundary corresponding to the currently associated objects from the scene boundary as the target scene boundary corresponding to the currently newly added objects; locally aligning the currently newly added objects based on the local coordinate system corresponding to the target scene boundary and the current associated objects, and obtaining the target newly added objects; placing the target newly added objects in the current indoor scene, and obtaining the target indoor scene. The use of this method can improve the accuracy of indoor scene generation.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, computer-readable storage medium and computer program product for generating an indoor scene based on scene boundaries. Background Art

[0002] With the rise of AIGC (Artificial Intelligence Generated Content) technology, deep learning models are widely used in various generation tasks, such as generating images, generating three-dimensional models, generating music, and generating scenes.

[0003] Existing scene generation methods, such as indoor scene generation for a room, need to consider the selection and layout of furniture. Generally, relevant furniture (such as beds and bedside tables) are extracted from the data set, and then the layout of the furniture is iteratively optimized to generate the final result. However, in the process of furniture layout, this method is limited in its generalization ability to generate diverse indoor scenes due to the lack of diverse indoor scene data. It usually has a good generation effect under conventional rectangular room boundaries, but there are large errors in the furniture layout in the indoor scenes generated under unconventional general room boundaries (such as polygonal room boundaries), resulting in low accuracy in the generation of indoor scenes. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, computer equipment, computer-readable storage medium and computer program product for generating indoor scenes based on scene boundaries, which can improve the generation accuracy of indoor scenes, in order to address the above technical problems.

[0005] In a first aspect, the present application provides a method for generating an indoor scene based on a scene boundary, comprising:

[0006] Obtaining a scene boundary and a current indoor object of a current indoor scene, predicting a new object in the current indoor scene based on the scene boundary and the current indoor object, and obtaining the current new object;

[0007] Based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects;

[0008] Based on the local coordinate system corresponding to the target scene boundary and the current associated object, the current newly added object is locally aligned to obtain the target newly added object;

[0009] Place the target new object in the current indoor scene to obtain the target indoor scene.

[0010] In a second aspect, the present application also provides an indoor scene generation device based on scene boundaries, comprising:

[0011] An object prediction module is used to obtain a scene boundary and a current indoor object of a current indoor scene, and predict a new object in the current indoor scene based on the scene boundary and the current indoor object to obtain the current new object;

[0012] A boundary determination module, for determining, based on the object layout features corresponding to the current indoor objects, the current associated objects associated with the current newly added objects in the current indoor objects, and obtaining the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects;

[0013] A local alignment module is used to locally align the current newly added object based on the local coordinate system corresponding to the target scene boundary and the current associated object to obtain the target newly added object;

[0014] The scene generation module is used to place target new objects in the current indoor scene to obtain a target indoor scene.

[0015] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0016] Obtaining a scene boundary and a current indoor object of a current indoor scene, predicting a new object in the current indoor scene based on the scene boundary and the current indoor object, and obtaining the current new object;

[0017] Based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects;

[0018] Based on the local coordinate system corresponding to the target scene boundary and the current associated object, the current newly added object is locally aligned to obtain the target newly added object;

[0019] Place the target new object in the current indoor scene to obtain the target indoor scene.

[0020] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the following steps are implemented:

[0021] Obtaining a scene boundary and a current indoor object of a current indoor scene, predicting a new object in the current indoor scene based on the scene boundary and the current indoor object, and obtaining the current new object;

[0022] Based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects;

[0023] Based on the local coordinate system corresponding to the target scene boundary and the current associated object, the current newly added object is locally aligned to obtain the target newly added object;

[0024] Place the target new object in the current indoor scene to obtain the target indoor scene.

[0025] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0026] Obtaining a scene boundary and a current indoor object of a current indoor scene, predicting a new object in the current indoor scene based on the scene boundary and the current indoor object, and obtaining the current new object;

[0027] Based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects;

[0028] Based on the local coordinate system corresponding to the target scene boundary and the current associated object, the current newly added object is locally aligned to obtain the target newly added object;

[0029] Place the target new object in the current indoor scene to obtain the target indoor scene.

[0030] The above-mentioned indoor scene generation method, device, computer equipment, computer-readable storage medium and computer program product based on scene boundary predict the current newly added objects in the current indoor scene according to the scene boundary of the current indoor scene and the current indoor objects, and can accurately predict the current newly added objects according to the global information of the current indoor scene, thereby realizing the global placement of the current newly added objects in the current indoor scene; then determine the current associated objects according to the object layout characteristics of the current indoor objects, thereby ensuring the layout rationality between the current newly added objects and the current associated objects; then use the local scene boundary corresponding to the current associated objects in the scene boundary as the local scene boundary of the current newly added objects. The target scene boundary is locally aligned with the current newly added objects according to the local coordinate system of the target scene boundary and the current associated objects, thereby realizing the accurate layout of the current newly added objects in the local scene of the target scene boundary, obtaining the target newly added objects, and improving the accuracy of the object layout of the target indoor scene, thereby improving the accuracy of the indoor scene generation; further, by accurately laying out the local scene of the target scene boundary in the scene boundary, the splitting processing of the local scene boundary in the scene boundary is realized without considering the overall scene boundary of the indoor scene, thereby improving the generalization ability of the indoor scene generation, and then ensuring the diversity of the indoor scene generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 An application environment diagram of an indoor scene generation method based on scene boundaries in one embodiment;

[0033] Figure 2 is a flow chart of a method for generating an indoor scene based on scene boundaries in one embodiment;

[0034] Figure 3 is a flow chart of a step of generating an indoor scene based on a scene boundary in one embodiment;

[0035] Figure 4 A schematic diagram of a process of generating a room scene in one embodiment;

[0036] Figure 5 is a structural block diagram of an indoor scene generation device based on scene boundaries in one embodiment;

[0037] Figure 6 is an internal structure diagram of a computer device in one embodiment;

[0038] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in another embodiment. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0040] The indoor scene generation method based on scene boundary provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, the terminal 102 communicates with the server 104 via a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed on the cloud or other network servers.

[0041] In one embodiment, the server 104 can obtain the scene boundary and the current indoor object of the current indoor scene through the terminal 102, predict the newly added object in the current indoor scene based on the scene boundary and the current indoor object, and obtain the currently newly added object; the server 104 determines the currently associated object associated with the currently added object in the current indoor object based on the object layout features corresponding to the current indoor object, and obtains the local scene boundary corresponding to the currently associated object from the scene boundary as the target scene boundary corresponding to the currently added object; the server 104 locally aligns the currently added object based on the local coordinate system corresponding to the target scene boundary and the currently associated object, and obtains the target newly added object; the server 104 places the target newly added object in the current indoor scene, and obtains the target indoor scene; and then the server 104 can return the target indoor scene to the terminal 102. Among them, the terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0042] In one embodiment, Figure 2 As shown, a method for generating an indoor scene based on a scene boundary is provided. This embodiment takes the method applied to a server as an example. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0043] Step 202: Acquire the scene boundary and current indoor objects of the current indoor scene, predict the newly added objects in the current indoor scene based on the scene boundary and the current indoor objects, and obtain the current newly added objects.

[0044] The current indoor scene refers to the indoor space where scene generation can be performed, such as a room in a house, a warehouse, a shopping mall, etc. The scene boundary refers to the physical boundary of the current indoor scene, such as the wall boundary of a room, etc. The current indoor object refers to the objects arranged in the current indoor scene, and the indoor object refers to the facilities in the indoor scene, such as furniture facilities, kitchen and bathroom facilities, commodity shelf facilities, etc. The current indoor object can be one or more. The current newly added object refers to the newly added object to be arranged in the current indoor scene.

[0045] Specifically, the server may receive a scene generation instruction for the current indoor scene sent by the terminal, and the scene generation instruction may carry scene data corresponding to the current indoor scene. In response to the scene generation instruction, the server identifies and obtains the scene boundary and the current indoor object from the scene data of the current indoor scene, including obtaining the attribute information corresponding to the current indoor object, including the object type, size and current position of the current indoor object in the global coordinate system of the current indoor scene. Then the server extracts features of the scene boundary and the attribute information of the current indoor object to obtain scene boundary features and object attribute features, and predicts the newly added objects in the current indoor scene based on the scene boundary features and the object attribute features, which may specifically be to predict the attribute information of the newly added objects in the current indoor scene to obtain the current newly added objects.

[0046] In one embodiment, the current indoor scene may be an empty scene without any current indoor objects, that is, the server identifies and obtains the scene boundary from the scene data corresponding to the current indoor scene, and extracts features of the obtained scene boundary to obtain scene boundary features. The server predicts the newly added objects in the current indoor scene based on the scene boundary features, which may be inputting the scene boundary features into a pre-trained newly added object attribute prediction model, obtaining the attribute information of the newly added objects in the current indoor scene output by the model, and obtaining the current newly added objects.

[0047] Step 204, based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects.

[0048] Among them, the object layout feature is used to characterize the object type feature of the indoor objects with layout association relationship. The layout association relationship can be the functional association relationship, position association relationship, appearance association relationship, etc. of the indoor objects. The layout association relationship type corresponding to the object layout feature can be determined according to the scene type of the current indoor scene. The current associated object refers to the current indoor object that has a layout association relationship with the current newly added object. The target scene boundary refers to the local scene boundary in the current indoor scene used to place the current newly added object.

[0049] Specifically, the server is pre-set with object layout feature sets corresponding to the scene types of each indoor scene, and the object layout feature sets include object layout features corresponding to each indoor object in the corresponding indoor scene. For example, the furniture layout feature set corresponding to the bedroom includes furniture layout features corresponding to the wardrobe and dressing table.

[0050] The server obtains the corresponding object layout feature set according to the scene type of the current indoor scene, and then obtains the object layout features corresponding to the current indoor object from the object layout feature set. The server performs feature matching between the object layout features corresponding to the current newly added object and the current indoor object, determines the successfully matched current indoor object as the current associated object associated with the current newly added object, and then obtains the local scene boundary corresponding to the current associated object from the scene boundary, indicating the local scene boundary where the current associated object is placed in the current indoor scene. The server uses the local scene boundary corresponding to the current associated object as the target scene boundary corresponding to the current newly added object.

[0051] In one embodiment, if the current indoor scene is an empty scene without the current indoor object, the server may determine the target scene boundary in each local scene boundary of the scene boundary corresponding to the current indoor scene according to the current position of the newly added object in the current indoor scene and the boundary information of the scene boundary corresponding to the current indoor scene. The boundary information may include the boundary position and boundary length of each local scene boundary in the scene boundary.

[0052] Step 206 , based on the local coordinate system corresponding to the target scene boundary and the current associated object, locally align the current newly added object to obtain the target newly added object.

[0053] Among them, local alignment refers to unifying the layout of the current newly added objects according to the layout of the current associated objects in the local scene corresponding to the boundary of the target scene, which can be understood as optimizing the layout of the current newly added objects so that the boundary of the current newly added objects coincides with the boundary of the target scene, and the placement angle of the current newly added objects is guaranteed to be consistent with the placement angle of the current associated objects. The target newly added object refers to the newly added object that has a unified layout with the current associated object after the local alignment of the current newly added object.

[0054] Specifically, the attribute information corresponding to the current newly added object includes the current position of the current newly added object in the global coordinate system of the current indoor scene. The server needs to align the coordinate systems of the current newly added object and the current associated object when performing local alignment on the current newly added object.

[0055] After determining the target scene boundary corresponding to the current newly added object, the server establishes a local coordinate system corresponding to the target scene boundary, and performs coordinate transformation on the current position and placement angle corresponding to the current newly added object and the current associated object according to the local coordinate system corresponding to the target scene boundary. The server extracts features from the target scene boundary, the attribute information corresponding to the current newly added object and the current associated object after coordinate transformation, and inputs the extracted features into a pre-trained attribute optimization model to obtain the optimized attribute information corresponding to the current newly added object, which may include the local position, object type, size, and placement angle of the current newly added object after attribute optimization in the local coordinates. The server obtains the target newly added object according to the optimized attribute information corresponding to the current newly added object.

[0056] Step 208: placing the target newly added object in the current indoor scene to obtain the target indoor scene.

[0057] Specifically, the server converts the local position and placement angle corresponding to the target newly added object to the target position and target placement angle in the global coordinate system of the current indoor scene, and places the target newly added object in the current indoor scene according to the target position and target placement angle to obtain the target indoor scene.

[0058] In one embodiment, an object database is pre-set in the server, including object sets of different types and sizes. After obtaining the optimized attribute information corresponding to the target newly added object, the server uses the object type and size in the optimized attribute information as matching conditions, selects objects that meet the matching conditions in the object database, obtains the target newly added object, and places the target newly added object in the current indoor scene according to the target position and target placement angle to obtain the target indoor scene.

[0059] In one embodiment, the server places the target newly added object in the current indoor scene according to the target position and the target placement angle, and obtains an updated scene corresponding to the current indoor scene. The placed target newly added object and the current indoor object can be used as existing objects in the updated scene corresponding to the current indoor scene. The server detects that the updated scene corresponding to the current indoor scene does not meet the scene generation end condition, such as the number condition of existing objects in the scene, the scene area condition in the scene, the type condition of existing objects in the scene, etc. The server predicts the next current newly added object in the current indoor scene based on the existing objects and scene boundaries in the updated scene corresponding to the current indoor scene, and enters the step of determining the current associated object associated with the current newly added object in the current indoor object based on the object layout feature corresponding to the current indoor object, until the updated scene corresponding to the current indoor scene meets the scene generation end condition, and the target indoor scene is obtained.

[0060] In the above-mentioned indoor scene generation method based on scene boundary, by predicting the current newly added objects in the current indoor scene according to the scene boundary of the current indoor scene and the current indoor objects, the current newly added objects can be accurately predicted according to the global information of the current indoor scene, and the global placement of the current newly added objects in the current indoor scene is realized; then the current associated objects are determined according to the object layout characteristics of the current indoor objects, and the layout rationality between the current newly added objects and the current associated objects is guaranteed; then the local scene boundary corresponding to the current associated objects in the scene boundary is used as the target scene boundary of the current newly added objects, and the current newly added objects and the current associated objects are locally aligned according to the local coordinate system of the target scene boundary, and the accurate layout of the current newly added objects in the local scene of the target scene boundary is realized, and the target newly added objects are obtained, thereby improving the accuracy of the object layout of the target indoor scene, thereby improving the accuracy of the indoor scene generation; further, by accurately arranging the local scene of the target scene boundary in the scene boundary, the splitting processing of the local scene boundary in the scene boundary is realized without considering the overall scene boundary of the indoor scene, thereby improving the generalization ability of indoor scene generation, and then ensuring the diversity of indoor scene generation.

[0061] In one embodiment, Figure 3 As shown, step 202 predicts new objects in the current indoor scene based on the scene boundary and the current indoor objects to obtain the current new objects, including:

[0062] Step 302, extracting features of the scene boundary and the current indoor object to obtain scene boundary features and object attribute features;

[0063] Step 304, inputting the scene boundary features and the object attribute features into the newly added object attribute prediction model to obtain attribute value distribution parameters corresponding to each object attribute of the newly added object;

[0064] Step 306, determining the target attribute value based on the attribute value distribution parameter, and obtaining the target attribute value corresponding to each object attribute;

[0065] Step 308: obtaining the currently added object based on the target attribute values ​​corresponding to the attributes of each object.

[0066] Among them, each object attribute is used to describe the object, including the object type, object size, current position and placement angle. The attribute value distribution parameter refers to the probability distribution parameter of each optional attribute value of the object attribute, which can be expressed as the attribute value distribution curve of the object attribute. The target attribute value refers to the attribute value of the currently selected object attribute.

[0067] Specifically, the server obtains the attribute information corresponding to the scene boundary and the current indoor object according to the scene data of the current indoor scene. The attribute information includes the attributes of each object corresponding to the current indoor object. The server extracts the features of the attribute information corresponding to the scene boundary and the current indoor object to obtain the scene boundary features and the object attribute features. Then, the pre-trained new object attribute prediction model is called, and the scene boundary features and the object attribute features are input into the new object attribute prediction model to obtain the attribute value distribution parameters corresponding to each object attribute of the new object. The server samples the attribute values ​​corresponding to each object attribute respectively, obtains the target attribute values ​​corresponding to each object attribute, and uses the target attribute values ​​corresponding to each object attribute as the attribute information corresponding to the current new object.

[0068] In a specific embodiment, the newly added object attribute prediction model can be an attention model for attribute prediction of newly added objects, which can be described as a global placement attention model, indicating that the current newly added object determined by the global placement attention model is a newly added object in the global coordinate system of the current indoor scene.

[0069] In the process of predicting new objects in the current indoor scene based on the scene boundary and the current indoor objects, the scene data of the current indoor scene can be represented as ,in, Represents the current set of indoor objects. A current indoor object can be represented as a three-dimensional bounding box ,in, Indicates the object category. is the 3D coordinate of the center of the 3D bounding box, Represents the three-dimensional size of the three-dimensional bounding box, Indicates the rotation angle of the 3D bounding box relative to the upward coordinate axis, that is, the placement angle. Represents the scene boundary, including multiple local scene boundaries, such as a collection of walls, a local scene boundary Contains the start and end points of its projection on the ground plane.

[0070] The server extracts features of the scene boundary and the current indoor objects to obtain the scene boundary feature S and object attribute feature Specifically, the server obtains a scene image of the current indoor scene in advance, and converts the scene boundary in the scene image into a two-dimensional boundary mask image. , the boundary mask image Input the pre-trained image encoder for encoding calculation to obtain the scene boundary feature S, which represents the feature encoding of the scene boundary B; and Input the pre-trained position encoder for encoding calculation to obtain the feature encoding of the object attribute features , The encoding method is shown in formula (1).

[0071] (1)

[0072] in, represents the learnable encoding layer for object categories, represents a fixed position code, Represents a concatenation operation.

[0073] The server encodes the scene boundary features S and object attribute features Generate an input sequence and set a sequence flag q to mark the elements in the input sequence. The server inputs the input sequence into the global placement attention model for calculation and outputs the attribute encoding result of the newly added object. , and encode the attribute result The input is sent to the decoder for decoding. The decoder consists of three MLPs (Multilayer Perceptrons). The decoder MLP outputs the attribute value distribution parameters corresponding to each object attribute of the newly added object and determines the target attribute value for it. The specific decoding method is shown in formula (2):

[0074] (2)

[0075] in, , , Respectively represent the predicted object category, current position and size in the global coordinate system of the current indoor scene; It represents the process of determining the target attribute value based on the attribute value distribution parameters corresponding to each object attribute of the newly added object.

[0076] Specifically, It can be represented by a one-hot vector, , It can be expressed using mixed logistic (continuous probability distribution) distribution, as shown in formula (3):

[0077] (3)

[0078] Indicates one of the three object properties of the newly added object: the current position, size or placement angle; , , They represent the weight, mean and variance of the kth logistic distribution, i.e., the attribute value distribution parameters corresponding to each object attribute. k can be 10. represents the attribute value distribution curve determined by the attribute value distribution parameters corresponding to each object attribute, then formula (3) can be understood as the target attribute value of the object attribute is determined by the attribute value distribution curve. The server can sample the attribute value distribution curve determined by the attribute value distribution parameters corresponding to each object attribute, obtain the target attribute value corresponding to each object attribute, and use the target attribute value corresponding to each object attribute as the attribute information corresponding to the current newly added object.

[0079] In this embodiment, by inputting scene boundary features and object attribute features into the new object attribute prediction model, the attribute prediction of the new object can accurately predict the new object based on the global information of the current indoor scene, thereby ensuring the rationality of the object layout in the indoor scene.

[0080] In one embodiment, before obtaining the scene boundary of the current indoor scene and the current indoor object in step 202, the method for generating an indoor scene based on the scene boundary further includes:

[0081] Acquire a reference data set, the reference data set including reference indoor objects corresponding to each reference indoor scene;

[0082] In the same reference indoor scene, the reference indoor objects are clustered based on the reference positions of the reference indoor objects to obtain various clustered object combinations;

[0083] Counting the repetition rates of each cluster object combination, and determining each target object combination from each cluster object combination based on the repetition rate;

[0084] Based on the combination of each target object, each object layout feature is generated, and based on the layout features of each object, an object layout feature set is obtained.

[0085] The reference data set is the sampled data used to extract the object layout feature set. The reference indoor scene refers to an indoor scene that has completed the object layout and can provide object layout features. The reference indoor object refers to the indoor object with the optimal layout in the reference indoor scene. The reference position refers to the position of the reference indoor object in the reference indoor scene.

[0086] Specifically, the server obtains a reference data set, and clusters the reference indoor objects in the same reference indoor scene according to the reference positions of the reference indoor objects corresponding to each reference data scene in the reference data set, to obtain each cluster object combination. Generally, the reference indoor objects whose position distance from the reference position is less than a preset threshold are taken as cluster object combinations. Then the server counts the repetition rate of each cluster object combination, and determines the cluster object combination with a repetition rate greater than the preset threshold or the highest repetition rate as the target object combination to obtain each target object combination, for example, the cluster object combination of "bed + bedside table" and "bed + bedside lamp" with the highest repetition rate is determined as the target object combination. Then the server generates each object layout feature according to each target object combination, such as object layout feature 1 = {bed; bedside table}, object layout feature 2 = {bed; bedside lamp}, and then the server can merge each object layout feature to obtain an object layout feature set.

[0087] The server may also re-integrate each object layout feature to obtain each updated object layout feature, and merge each updated object layout feature according to each updated object layout feature to obtain an object layout feature set. Specifically, the same keywords may be extracted from each object layout feature, and the combined objects in the object layout features having the same keywords are determined, and the same keywords and combined objects are merged according to the weights of the determined combined objects. For example, the same keyword {bed} is extracted from object layout feature 1={bed; bedside table} and object layout feature 2={bed; bedside lamp}, and object layout feature 1 and object layout feature 2 are merged according to the weights of {bedside table} and {bedside lamp} to obtain the updated object layout feature {bed; bedside table; bedside lamp}.

[0088] In this embodiment, by counting the repetition rates of each cluster object combination, each target object combination is determined from each cluster object combination based on the repetition rate, and each object layout feature is generated according to each target object combination, so that the current associated object associated with the current newly added object can be subsequently determined according to the object layout features, thereby ensuring the rationality of the layout of the current newly added object and the current associated object.

[0089] In one embodiment, step 204, based on the object layout features corresponding to the current indoor objects, determining the current associated objects associated with the current newly added objects in the current indoor objects includes:

[0090] Finding object layout features corresponding to the current indoor objects from the object layout feature set, and determining the respective combined objects corresponding to the respective current indoor objects based on the found object layout features;

[0091] The current newly added object is matched with each combined object, and the current indoor object to which the successfully matched combined object belongs is determined as the current associated object associated with the current newly added object.

[0092] Specifically, the server matches the current indoor object with each object layout feature in the object layout feature set, obtains the object layout feature containing the current indoor object, and obtains the object layout feature corresponding to the current indoor object. Then the server obtains each combined object from the object layout feature corresponding to the current indoor object, matches the current newly added object with each combined object, and uses the current indoor object to which the successfully matched combined object belongs as the current associated object associated with the current newly added object. For example, the current indoor object is {wardrobe}, and the current newly added object is {full-length mirror}. The object layout feature of {wardrobe} found in the object layout feature set is {wardrobe; full-length mirror}, and the combined object {full-length mirror} contained therein matches the current newly added object {full-length mirror}. Then the current indoor object {wardrobe} is used as the current associated object associated with the current newly added object {full-length mirror}.

[0093] In this embodiment, by matching the current newly added object with each combined object, the current indoor object to which the successfully matched combined object belongs is determined as the current associated object associated with the current newly added object, thereby ensuring the accuracy of the current associated object and improving the rationality of the layout of the current newly added object and the current associated object.

[0094] In one embodiment, the method for generating an indoor scene based on a scene boundary further includes:

[0095] When the current newly added object fails to match each combined object, the target scene boundary corresponding to the current newly added object is determined from each local scene boundary based on the position distance between the current position corresponding to the current newly added object and each local scene boundary in the scene boundary.

[0096] Specifically, when the server fails to match the current newly added object with each combined object, it obtains the current position corresponding to the current newly added object, and calculates the position distance between the current position of the current newly added object and the local scene boundary, and determines the target scene boundary corresponding to the current newly added object in each local scene boundary according to the position distance. Generally, the local scene boundary with the shortest position distance is determined as the target scene boundary.

[0097] In a specific embodiment, when it is determined that there are multiple currently associated objects associated with the currently added object according to the object layout features corresponding to the current indoor object, the local scene boundaries corresponding to each currently associated object are obtained as candidate local scene boundaries. Then, the position distance between the current position of the currently added object and each candidate local scene boundary is calculated, and the candidate local scene boundary with the shortest position distance is determined as the target scene boundary corresponding to the currently added object.

[0098] In this embodiment, when the current newly added object fails to match each combined object, the local scene boundary with the shortest position clustering is used as the target scene boundary, so that the target scene boundary can be quickly determined, the local alignment efficiency of the current newly added object is improved, and the efficiency of indoor scene generation is thereby improved.

[0099] In one embodiment, step 206, based on the local coordinate system corresponding to the target scene boundary, locally aligning the current newly added object and the current associated object to obtain the target newly added object includes:

[0100] Based on the local coordinate system corresponding to the target scene boundary, coordinate transformation is performed on the newly added objects and the currently associated objects;

[0101] Extract features of the target scene boundary, the current newly added object after coordinate transformation, and the current associated object to obtain target scene boundary features, newly added object attribute features, and associated object attribute features;

[0102] Unifying the feature dimensions of the target scene boundary features, the newly added object attribute features, and the associated object attribute features, inputting the target scene boundary features, the newly added object attribute features, and the associated object attribute features with unified feature dimensions into the attribute optimization model to obtain the attribute optimization value corresponding to the newly added object attribute features;

[0103] The target new object is obtained based on the property optimization value corresponding to the property characteristics of the new object.

[0104] The attribute optimization value refers to the attribute value of each object attribute that can represent the optimal layout of the current newly added object in the local scene where the boundary of the target scene is located after the current newly added object is locally aligned.

[0105] Specifically, after determining the target scene boundary corresponding to the current newly added object, the server establishes a corresponding local coordinate system according to the target scene boundary, performs coordinate transformation on the attribute information corresponding to the current newly added object and the current associated object, and obtains the attribute information corresponding to the current newly added object and the current associated object after coordinate transformation. Generally, the coordinate transformation is performed on the current position and placement angle of the current newly added object and the current associated object, and the attribute information after coordinate transformation includes the local position and local placement angle corresponding to the current newly added object and the current associated object in the local coordinate system.

[0106] Then the server extracts the converted attribute information after the coordinate transformation corresponding to the target scene boundary, the current newly added object and the current associated object, and obtains the target scene boundary feature, the newly added object attribute feature and the associated object attribute feature. The target scene boundary feature can be the length feature of the target scene boundary. Since the target scene boundary feature has different feature dimensions from the newly added object attribute feature and the associated object attribute feature, that is, the number of features is different, it is necessary to unify the feature dimensions of the target scene boundary feature, the newly added object attribute feature and the associated object attribute feature. The feature dimension unification can be achieved by setting a fully connected layer when extracting the target scene boundary feature and the newly added object attribute feature. Then the server inputs the target scene boundary feature, the newly added object attribute feature and the associated object attribute feature with unified feature dimensions into the attribute optimization model, obtains the attribute optimization value corresponding to the newly added object attribute feature, and uses the attribute optimization value corresponding to the newly added object attribute feature as the optimized attribute information of the target newly added object. The server can place the target newly added object in the current indoor scene according to the optimized attribute information of the target newly added object.

[0107] In a specific embodiment, the attribute optimization model can be an attention model for local alignment, which can be described as a local alignment attention model. After the local coordinate system is converted, the target scene boundary features, the newly added object attribute features after the coordinate conversion, and the associated object attribute features are extracted. The server generates an input sequence based on the target scene boundary features, the newly added object attribute features after the coordinate conversion, and the associated object attribute features. The input sequence includes a boundary mark W corresponding to the target scene boundary features, an associated object mark W corresponding to the associated object attribute features, and an associated object mark W corresponding to the associated object attribute features. and an indicator mark c. The boundary mark W can be obtained by The encoding method of the boundary marker W is shown in formula (4):

[0108] (4), where FC represents the fully connected layer, which is used to unify the feature dimension.

[0109] Related object flag As shown in formula (5):

[0110] (5)

[0111] in, , They respectively represent the local position and local direction angle of the current associated object after coordinate transformation.

[0112] The indicator c is used to collect the attribute features of the newly added object of the current newly added object, which can be obtained by encoding and calculating the attribute features of the newly added object after coordinate transformation. The encoding method of the indicator c is shown in formula (6):

[0113] (6)

[0114] Among them, FC represents the fully connected layer, which is used to unify the feature dimension; It is the local position of the current newly added object in the local coordinate system in the attribute features of the newly added object after the coordinate transformation, that is, the center coordinates of the three-dimensional bounding box corresponding to the current newly added object after the coordinate transformation.

[0115] The server inputs the input sequence into the local alignment attention model to obtain the optimized attribute encoding result of the current newly added object , and then encode the optimization attribute of the newly added object The input is sent to the decoder for decoding. The decoder consists of four MLPs (Multilayer Perceptrons). The decoder MLP outputs the attribute optimization value distribution parameters corresponding to each object attribute of the current newly added object and determines the attribute optimization value for it. The specific decoding method is shown in formula (7):

[0116] (7)

[0117] in, , , , They respectively represent the property optimization values ​​corresponding to the object category, current position, size and placement angle of the newly added object in the local coordinate system; It represents the process of determining the target attribute value based on the attribute optimization value distribution parameters corresponding to each object attribute of the current newly added object.

[0118] In this embodiment, by performing coordinate transformation on the current newly added object and the current associated object according to the local coordinate system corresponding to the target scene boundary, and inputting the target scene boundary features, the newly added object attribute features and the associated object attribute features after the coordinate transformation and feature dimension unification into the attribute optimization model, the attribute optimization value corresponding to the newly added object attribute features is obtained, which can realize the accurate layout of the local scene of the current newly added object, ensure the layout accuracy of the target newly added object, and thus improve the accuracy of indoor scene generation.

[0119] In one embodiment, step 208, placing a target newly added object in the current indoor scene to obtain a target indoor scene, includes:

[0120] Place the target new object in the current indoor scene to obtain an intermediate indoor scene;

[0121] The intermediate indoor scene is taken as the current indoor scene, and the set of the current indoor objects and the target newly added objects is taken as the current indoor objects. The process returns to the step of predicting the newly added objects based on the scene boundary and the current indoor objects, and the step of obtaining the current newly added objects is executed until the end condition is met to obtain the target indoor scene.

[0122] Specifically, the server places the target new object in the current indoor scene according to the optimized attribute information of the target new object, and obtains an intermediate indoor scene, which represents the updated scene of the current indoor scene after the target new object is newly arranged. Then the server uses the intermediate indoor scene as the current indoor scene, and the set of the current indoor objects and the target new objects as the current indoor objects, and returns to the step of predicting the new objects based on the scene boundary and the current indoor objects to obtain the current new objects, until the end condition is met to obtain the target indoor scene. The end condition can be that when the new object attribute prediction model outputs an end mark, it means that the current indoor scene no longer needs to add new objects, and the server ends the indoor scene generation and uses the current indoor scene as the target indoor scene.

[0123] In this embodiment, by adding the predicted current newly added object to the existing objects in the current indoor scene, the next newly added object is predicted autoregressively until the end condition is met, and the target indoor scene is obtained, thereby realizing the automatic generation of the indoor scene and improving the efficiency of indoor scene generation.

[0124] In a specific embodiment, the indoor scene generation method based on scene boundaries can be applied to an indoor scene generation system, which includes a global placement module, a boundary allocation module, and a local alignment module. The global placement module includes a global attention model for predicting the current newly added objects;

[0125] The boundary allocation module is used to determine the current associated object associated with the current newly added object. It is set with co-occurrence rules and distance standards. The co-occurrence rule is a rule used to determine whether a group of objects should be assigned to the same local scene boundary. The current associated object can be determined based on the object layout characteristics; the distance standard refers to selecting the local scene boundary closest to the current newly added object. Objects that meet the co-occurrence rule and distance standard can be determined as the current associated object, that is, the current newly added object is assigned to the local scene boundary that is closest and where the current associated object is placed.

[0126] The local alignment module includes a local alignment attention model, which is used to locally align the current newly added objects.

[0127] In a specific embodiment, Figure 4As shown, a schematic diagram of the process of room scene generation is provided. The indoor scene generation system can generate the furniture layout scene in the room based on the wall boundary of the room. The specific steps are as follows:

[0128] The current room scene is input into the global placement module, and the wall boundary and current furniture information in the current room scene are extracted through the global placement model (such as Figure 4 The wall boundary is input into the boundary mask encoder (i.e., the pre-trained image encoder) to obtain the wall boundary feature S, and the current furniture information is input into the coarse object encoder (i.e., the pre-trained position encoder, which is used to encode the current furniture information at the coarse position) to obtain the furniture features of the three existing furniture. , , , and set a sequence flag q to convert the room boundary feature S, furniture feature , , and sequence flag q as the input sequence to the global placement attention module, and output the attribute encoding result of the newly added object Then, a coarse furniture decoder (i.e., a decoder consisting of three MLPs) is used to encode the attributes of the newly added objects. Decode and obtain the rough furniture prediction result, that is, the target attribute value corresponding to each object attribute of the newly added furniture.

[0129] The boundary allocation module selects candidate wall boundaries that meet the co-occurrence rule among the existing furniture and all wall boundaries, and then determines the target wall boundary among the candidate wall boundaries according to the distance standard, and allocates the current newly added furniture to the target wall boundary. If no existing furniture and wall boundaries meet the co-occurrence rule, the target wall boundary is determined from all wall boundaries according to the distance standard. Figure 4 middle , , Represents three existing furniture.

[0130] The local alignment module constructs a local coordinate system according to the target wall boundary, and encodes the length of the target wall boundary to obtain the wall mark W. Then, the coordinate transformation module transforms the coordinates of the newly added furniture and the associated furniture, and the alignment object encoder transforms the associated furniture after the coordinate transformation. Encode and get the associated furniture Encoding features in local coordinates , and generate an indicator mark c according to the target attribute value of each object attribute predicted by the global placement attention module, and convert the wall mark W and the encoding feature and the indicator c as the input sequence, which is input to the local alignment attention module. The local alignment attention module outputs the optimized attribute encoding result of the current newly added furniture. , using the aligned furniture decoder (i.e., a decoder consisting of four MLPs) to encode the optimized attributes of the newly added furniture Decode and get the detailed furniture prediction results, that is, the property optimization values ​​corresponding to each object attribute of the target new furniture. Then return to the global placement model, autoregressively generate and predict the next new furniture, until the global placement model predicts the end of the sequence mark indicating the end <eos>.

[0131] In this embodiment, by predicting the current newly added object in the current indoor scene according to the scene boundary of the current indoor scene and the current indoor object, the current newly added object can be accurately predicted according to the global information of the current indoor scene, and the global placement of the current newly added object in the current indoor scene is realized; then the current associated object is determined according to the object layout characteristics of the current indoor object, and the layout rationality between the current newly added object and the current associated object is ensured; then the local scene boundary corresponding to the current associated object in the scene boundary is used as the target scene boundary of the current newly added object, and the current newly added object and the current associated object are locally aligned according to the local coordinate system of the target scene boundary, so as to realize the accurate layout of the current newly added object in the local scene of the target scene boundary, obtain the target newly added object, improve the accuracy of the object layout of the target indoor scene, and thus improve the accuracy of the indoor scene generation; further, by accurately laying out the local scene of the target scene boundary in the scene boundary, the splitting processing of the local scene boundary in the scene boundary is realized without considering the overall scene boundary of the indoor scene, thereby improving the generalization ability of indoor scene generation, and then ensuring the diversity of indoor scene generation.

[0132] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0133] Based on the same inventive concept, the embodiment of the present application also provides a scene boundary-based indoor scene generation device for implementing the scene boundary-based indoor scene generation method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more scene boundary-based indoor scene generation device embodiments provided below can refer to the limitations of the scene boundary-based indoor scene generation method above, and will not be repeated here.

[0134] In an exemplary embodiment, Figure 5 As shown, an indoor scene generation device 500 based on scene boundaries is provided, comprising an object prediction module 502, a boundary determination module 504, a local alignment module 506 and a scene generation module 508, wherein:

[0135] The object prediction module 502 is used to obtain the scene boundary and the current indoor object of the current indoor scene, and predict the newly added object in the current indoor scene based on the scene boundary and the current indoor object to obtain the current newly added object;

[0136] The boundary determination module 504 is used to determine the current associated object associated with the current newly added object in the current indoor objects based on the object layout features corresponding to the current indoor objects, and obtain the local scene boundary corresponding to the current associated object from the scene boundary as the target scene boundary corresponding to the current newly added object;

[0137] A local alignment module 506 is used to locally align the current newly added object based on the local coordinate system corresponding to the target scene boundary and the current associated object to obtain the target newly added object;

[0138] The scene generation module 508 is used to place the target new object in the current indoor scene to obtain the target indoor scene.

[0139] In one embodiment, the object prediction module 502 is further used to extract features from the scene boundary and the current indoor object to obtain scene boundary features and object attribute features; input the scene boundary features and the object attribute features into a newly added object attribute prediction model to obtain attribute value distribution parameters corresponding to each object attribute of the newly added object; determine target attribute values ​​based on the attribute value distribution parameters to obtain target attribute values ​​corresponding to each object attribute; and obtain the current newly added object based on the target attribute values ​​corresponding to each object attribute.

[0140] In one embodiment, the scene boundary-based indoor scene generation device 500 is further used to obtain a reference data set, wherein the reference data set includes reference indoor objects corresponding to each reference indoor scene; in the same reference indoor scene, the reference indoor objects are clustered based on their reference positions to obtain each clustered object combination; the repetition rate of each clustered object combination is counted, and each target object combination is determined from each clustered object combination based on the repetition rate; each object layout feature is generated based on each target object combination, and an object layout feature set is obtained based on each object layout feature.

[0141] In one embodiment, the boundary determination module 504 is further used to search for the object layout features corresponding to the current indoor object from the object layout feature set, and determine the respective combined objects corresponding to the respective current indoor objects based on the found object layout features; match the current newly added object with the respective combined objects, and determine the current indoor object to which the successfully matched combined object belongs as the current associated object associated with the current newly added object.

[0142] In one embodiment, the scene boundary-based indoor scene generation device 500 is also used to determine the target scene boundary corresponding to the current newly added object from the local scene boundaries based on the position distance between the current position corresponding to the current newly added object and the local scene boundaries in the scene boundary when the current newly added object fails to match with the various combined objects.

[0143] In one embodiment, the scene generation module 508 is further used to perform coordinate transformation on the current newly added object and the current associated object based on the local coordinate system corresponding to the target scene boundary; perform feature extraction on the target scene boundary, the current newly added object after coordinate transformation, and the current associated object to obtain target scene boundary features, newly added object attribute features, and associated object attribute features; perform feature dimension unification on the target scene boundary features, the newly added object attribute features, and the associated object attribute features, and input the target scene boundary features, the newly added object attribute features, and the associated object attribute features with unified feature dimensions into an attribute optimization model to obtain attribute optimization values ​​corresponding to the newly added object attribute features; and obtain the target newly added object based on the attribute optimization values ​​corresponding to the newly added object attribute features.

[0144] In one embodiment, the scene generation module 508 is further configured to place the target newly added object in the current indoor scene to obtain an intermediate indoor scene;

[0145] The intermediate indoor scene is taken as the current indoor scene, and the set of the current indoor objects and the target newly added objects is taken as the current indoor objects. The process returns to the step of predicting new objects based on the scene boundary and the current indoor objects to obtain the current newly added objects, and the process is executed until the end condition is met to obtain the target indoor scene.

[0146] Each module in the above-mentioned indoor scene generation device based on scene boundaries can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.

[0147] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store current indoor scene data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for generating an indoor scene based on a scene boundary is implemented.

[0148] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 7 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, near field communication (NFC) or other technologies. When the computer program is executed by the processor, a method for generating an indoor scene based on a scene boundary is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0149] Those skilled in the art will understand that Figure 6-Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0150] In an exemplary embodiment, in one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented.

[0151] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0152] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0153] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0154] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0155] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0156] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be construed as limiting the scope of the present application. It should be noted that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.< / eos>

Claims

1. A method for generating indoor scenes based on scene boundaries, characterized in that: The method comprises: Acquire a scene boundary and a current indoor object of a current indoor scene, perform feature extraction on the scene boundary and the current indoor object, and obtain scene boundary features and object attribute features; Inputting the scene boundary features and the object attribute features into a newly added object attribute prediction model to obtain attribute value distribution parameters corresponding to each object attribute of the newly added object; Determine the target attribute value based on the attribute value distribution parameter, and obtain the target attribute value corresponding to each of the object attributes; Obtaining the current newly added object based on the target attribute values ​​corresponding to the attributes of each object; Based on the object layout features corresponding to the current indoor objects, determine the current associated objects associated with the current newly added objects in the current indoor objects, and obtain the local scene boundary corresponding to the current associated objects from the scene boundary as the target scene boundary corresponding to the current newly added objects; Based on the local coordinate system corresponding to the target scene boundary and the current associated object, locally align the current newly added object to obtain the target newly added object; The target newly added object is placed in the current indoor scene to obtain a target indoor scene.

2. The method according to claim 1, characterized in that Before obtaining the scene boundary of the current indoor scene and the current indoor object, the method further includes: Acquire a reference data set, wherein the reference data set includes reference indoor objects corresponding to each reference indoor scene; In the same reference indoor scene, clustering the reference indoor objects based on the reference positions of the reference indoor objects to obtain clustered object combinations; Counting the repetition rates of the various clustered object combinations, and determining various target object combinations from the various clustered object combinations based on the repetition rates; Each object layout feature is generated based on the combination of each target object, and an object layout feature set is obtained based on the layout features of each object.

3. The method according to claim 2, characterized in that The determining, based on the object layout features corresponding to the current indoor objects, the current associated objects associated with the current newly added objects in the current indoor objects includes: Searching for object layout features corresponding to the current indoor objects from the object layout feature set, and determining the respective combined objects corresponding to the respective current indoor objects based on the searched object layout features; The current newly added object is matched with each of the combined objects, and the current indoor object to which the successfully matched combined object belongs is determined as the current associated object associated with the current newly added object.

4. The method according to claim 3, characterized in that The method further comprises: When the current newly added object fails to match the respective combined objects, the target scene boundary corresponding to the current newly added object is determined from the respective local scene boundaries based on the position distance between the current position corresponding to the current newly added object and the respective local scene boundaries in the scene boundary.

5. The method according to claim 1, characterized in that The locally aligning the current newly added object based on the local coordinate system corresponding to the target scene boundary and the current associated object to obtain the target newly added object includes: Based on the local coordinate system corresponding to the boundary of the target scene, coordinate transformation is performed on the current newly added object and the current associated object; Extracting features of the target scene boundary, the current newly added object after coordinate transformation, and the current associated object to obtain target scene boundary features, newly added object attribute features, and associated object attribute features; Unifying the feature dimensions of the target scene boundary feature, the newly added object attribute feature, and the associated object attribute feature, and inputting the target scene boundary feature, the newly added object attribute feature, and the associated object attribute feature with unified feature dimensions into an attribute optimization model to obtain an attribute optimization value corresponding to the newly added object attribute feature; A target new object is obtained based on the attribute optimization value corresponding to the attribute feature of the new object.

6. The method according to claim 5, characterized in that The step of obtaining a target new object based on the property optimization value corresponding to the property feature of the new object includes: The attribute optimization value corresponding to the attribute feature of the newly added object is used as a matching condition, and objects that meet the matching condition are screened in a preset object database to obtain the target newly added object.

7. The method according to claim 1, characterized in that Placing the target newly added object in the current indoor scene to obtain a target indoor scene includes: Placing the target newly added object in the current indoor scene to obtain an intermediate indoor scene; The intermediate indoor scene is taken as the current indoor scene, and the set of the current indoor objects and the target newly added objects is taken as the current indoor objects. The process returns to the step of extracting features of the scene boundary and the current indoor objects to obtain scene boundary features and object attribute features, and the process is executed until an end condition is met to obtain a target indoor scene.

8. A device for generating indoor scenes based on scene boundaries, characterized in that: The device comprises: The object prediction module is used to obtain the scene boundary and the current indoor object of the current indoor scene, perform feature extraction on the scene boundary and the current indoor object to obtain scene boundary features and object attribute features; input the scene boundary features and the object attribute features into the newly added object attribute prediction model to obtain attribute value distribution parameters corresponding to each object attribute of the newly added object; determine the target attribute value based on the attribute value distribution parameters to obtain the target attribute values ​​corresponding to each object attribute; obtain the current newly added object based on the target attribute values ​​corresponding to each object attribute; a boundary determination module, configured to determine, in the current indoor objects, a current associated object associated with the current newly added object based on the object layout features corresponding to the current indoor object, and obtain, from the scene boundary, a local scene boundary corresponding to the current associated object as a target scene boundary corresponding to the current newly added object; A local alignment module, configured to locally align the current newly added object based on the local coordinate system corresponding to the target scene boundary and the current associated object, to obtain a target newly added object; The scene generation module is used to place the target newly added object in the current indoor scene to obtain a target indoor scene.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images